[evlatests] Phase jumps on EA14

[Michael Rupen]

> We are working on this from the pipeline side, but at the same time, such 
> things need to be fixed from the technical sided as well.  The pipeline is 
> part of end-to-end operations now.

Which might be one reason to apply the same considerations to on-line as to
off-line calibration ;)

>  It really does try to do something 
> sensible in picking a refant -- it is based on location in the array and the 
> amount of unflagged data (to avoid picking an antenna that is shadowed or 
> missing a receiver).

Of course this is true, and I didn't mean to say that the current
choice was not rational. Basically I'm advocating adding another sensible 
criterion to the list, based on global stability, which the pipeline
fundamentally cannot know from internal information in the data set at 
present.  I would like to add that long-term information eventually to the SDM 
so that we can use it in post-processing, as we do in the real-time system. 
Of course TelCal can and should learn from the post-processing approach as 
well, though there I'd say the major issue is the robustness of the algorithm 
to bad data, rather than the choice of reference antenna.

>  But it cannot test for phase jumps.  I don't think 
> using the same on as Telcal is always going to work either.  We are working 
> on providing a list of antennas that it should NOT use.

This seems a much harder problem to me than an ordered list of
ones which we track carefully and believe *do* work.  Such a list can be
as short as we want it to be -- we only need a few truly trustworthy
dishes, after all -- and even better, that list already exists. We 
cannot be in the mode of requiring that all antennas work perfectly at all 
bands, or even requiring that we know immediately when any antenna/band/IF 
combo goes bad -- the system is too large and complex for that, and we do not 
have the manpower to make it as gloriously tracked as we'd like it to be. 
Even if _we_ could, presumably we are aiming at this pipeline being extensible 
to other and larger observatories -- at least ALMA, and I would hope others
if we do a good job -- and those are going to be even more complex and
harder to track.


              Michael